Home · Blog · Translating messages in a customer chat: how to answer foreigners without knowing their language
Translating messages in a customer chat: how to answer foreigners without knowing their language
A customer writes in Kazakh, Uzbek, English or Turkish, and the operator only knows Russian. It used to mean copying the text into a translator, translating the reply back and hoping nothing got lost. In CHBT Conversations translation is built right into the chat: an incoming message is translated with one button, and the typed reply — into the customer’s language. Here is how it works, what it costs and where the limits are.
When a business needs translation in chat
- Sales in the CIS. A store serves Kazakhstan, Uzbekistan, Kyrgyzstan — some customers write in their native language.
- Tourism and hotels. Guests from different countries ask about check-in, transfers, bookings.
- Marketplaces and export. A buyer asks about a product in English or Chinese.
- Service and delivery in big cities. Customers find it easier to explain a problem in their own language.
In all these cases the operator does not need to know the language — it is enough to understand what was asked and answer so the customer understands.
How to translate an incoming message
An incoming message in Conversations has a Translate button. Press it — Translating… appears, and a second later the translation into the dashboard language shows under the message: Russian, or English if the dashboard is open in English.
The customer’s language is detected automatically, you do not need to choose it. Only you and your team see the translation — nothing is sent to the customer. If the translation fails, you will see Translation failed — just press again.
How to reply in the customer’s language
- Write the reply in Russian in the reply field as usual.
- Press To customer’s language. The button hint: “Translate the typed reply into the customer’s language — check it and send”.
- The text in the field is replaced with the translation. Check it: names, prices, addresses, links.
- Send the message yourself.
CHBT detects the customer’s language from their last incoming text message. If the customer has not written any text yet — for example, they only tapped buttons or sent a photo — you will see “Could not detect the customer’s language — there are no text messages from them.” In that case ask the customer to write a couple of words or reply in Russian.
Why the translation is not sent automatically
There is deliberately no automatic sending of translations. Machine translation is good for everyday chat but sometimes gets names, terms and nuances wrong: a “booking” can turn into “armour”, and a polite form into a familiar one. The operator sees the translation before sending and can correct it or rewrite the original more simply.
Tip: write the original reply in short simple sentences without slang or abbreviations — the translation will be more accurate.
Cost and plan
Translation works on plans with AI — Basic and Business, and also in the trial: 3 days of Business turn on by themselves when you connect the first channel. Each translation — of an incoming message or a reply — costs 1 AI credit. Up to 4,000 characters are translated at a time.
If AI credits run out, translation becomes unavailable until you top up. The credit balance is shown in the dashboard; how many credits AI agent replies and translation use is described on the pricing page.
Where the data is processed
Translation is done by Yandex Translate through a Yandex Cloud service account — the same one CHBT uses for YandexGPT and voice message recognition. Conversation data is processed in Russia. For a business that follows Federal Law 152, this matters: the customer’s text does not go to foreign services. More about personal data in a bot is in the article “152-FZ and chatbot”.
Translation and the AI agent
Translation in Conversations is an operator tool: it helps the person running the chat. If an AI agent runs the conversation, separate translation is usually not needed: modern models understand questions in many languages and reply in the customer’s language. Add something like language: reply in the customer’s language to the agent’s Rules by topic, and check it in the test chat in 2–3 languages. How to set up the agent is covered in the article “How to set up an AI agent”.
When the agent hands a complex question to a person, the operator continues the conversation with translation in Conversations.
Practice for the team
- Quick replies in Russian. Keep Quick replies in Russian and translate them with To customer’s language — this way the texts stay the same for the whole team.
- Notes in your own language. Notes are visible only to the team, there is no need to translate them.
- Check numbers and links. Translation does not change numbers, but check that the date format and currency are clear to the customer.
- Complex cases to a native speaker. A contract, a claim, medical questions — better hand them to an employee who knows the language.
Common mistakes
- Sending without checking. The translation replaces the text in the field — at least skim it.
- Long complex sentences. The simpler the original, the better the translation.
- Replying before the customer’s first message. Without incoming text the customer’s language cannot be detected.
- Translating “just in case”. Each translation costs 1 AI credit. Translate what you really do not understand.
Example: chatting with a customer from Uzbekistan
A customer writes to a store’s Telegram in Uzbek, asking about a size and delivery to Tashkent. Here is how it looks for the operator:
- The message arrives in Conversations. The operator presses Translate and reads the Russian translation: “Hello, do you have this jacket in size M and how long is delivery to Tashkent?”.
- The operator checks stock and writes a reply in Russian: “Hello! Size M is available. Delivery to Tashkent takes 5–7 days, the cost is shown at checkout”.
- Presses To customer’s language — the text in the field is replaced with an Uzbek translation.
- Checks that the size and times are in place, and sends.
The whole cycle takes as long as an ordinary reply — without copying into an external translator and back.
Which languages are supported
Translation is done by Yandex Translate — it works with most common languages: English, Kazakh, Uzbek, Kyrgyz, Tajik, Armenian, Georgian, Azerbaijani, Turkish, Chinese, German, French, Spanish and many others. The customer’s language is detected automatically from the text of their message.
Incoming messages are translated into the dashboard language — Russian or English. If someone on the team works in the dashboard in English, they get the translation in English.
What is not translated
- Messages without text. Photos, files and stickers cannot be translated. On plans with AI, voice messages are transcribed to text — after transcription they can be translated like an ordinary message.
- Bot replies. The bot flow works in the language you wrote it in. If you have many customers in another language, make a separate flow branch or give the first line to an AI agent.
- Notes and quick replies. They stay in the language they were written in; a quick reply can be inserted into the field and translated with To customer’s language.
How to tell whether you need translation
Look in Conversations over the last month at how many chats were not in Russian. If these are isolated cases, built-in translation is enough and you do not need a separate employee who knows the language. If the share is noticeable — 10–20% or more — consider a flow branch in that language, an AI agent with a rule to reply in the customer’s language and, perhaps, a native-speaking employee for complex cases. Translation in Conversations bridges the gap between these decisions: you start replying right away, without waiting to hire.
Conversations and operatorsDistributing conversations between operatorsHow to set up an AI agentPlans